DerivedFunction / finbert-derivative-usage-classifier

Model Card for Derivative Disclosure Classifier

This is a text classification model fine-tuned to identify and classify sentences in financial reports that discuss derivative instruments. It is designed to work with text extracted from SEC filings like the 10-K.

Model Details

Model Description

This model is a fine-tuned version of bert-base-uncased intended for a specific financial NLP task: detecting disclosures related to derivatives. It processes sentences from corporate filings and classifies them, likely to determine if they contain substantive information about derivatives. The model is intended to be used as part of a larger data processing pipeline that first sources documents from the SEC EDGAR database, extracts relevant text chunks based on keywords, and then uses this model for fine-grained classification.

• Developed by: [More Information Needed]
• Model type: BERT-based text classifier
• Language(s) (NLP): English
• License: Apache-2.0
• Finetuned from model: google-bert/bert-base-uncased

Model Sources [optional]

• Repository: [More Information Needed]

Uses

Direct Use

The model is intended to be used via an API to classify individual sentences or short paragraphs extracted from financial documents. The primary use case is to automate the analysis of SEC filings to identify derivative-related statements.

Example: A user provides a batch of sentences, and the model returns a corresponding list of predictions (e.g., 1 for a derivative disclosure, 0 for not).

Downstream Use [optional]

This model can be a component in a larger financial analysis system. Potential downstream uses include:

• Aggregating sentence-level predictions to a document-level score to assess a company's overall derivative exposure.
• Tracking changes in derivative disclosures for a specific company over time.
• Building a large-scale database of derivative usage across an entire market sector.

Out-of-Scope Use

This model is highly specialized and should not be used for:

• General-purpose financial text classification.
• Analysis of documents outside of formal financial reporting, such as news articles, social media, or earnings call transcripts, as the language and structure are different.
• Making financial decisions without verification by a human expert. The model is a tool for screening and is not a substitute for professional financial analysis.
• Identifying derivative discussions that do not use the specific keywords the parent script was designed to search for.

Bias, Risks, and Limitations

The model's performance is subject to several limitations and potential biases stemming from its training data and methodology:

• Keyword-driven Bias: The training data is sourced by searching for a predefined list of keywords (e.g., "interest rate swap," "notional"). The model may therefore be biased towards recognizing these specific terms and could fail to identify disclosures that use synonymous or novel phrasing.
• Domain Specificity: The model is trained on the formal, structured language of U.S. SEC filings. Its performance on financial reports from other jurisdictions or document types is untested and likely to be significantly lower.
• Lack of Context: By classifying sentences or small chunks of text, the model may miss broader context from the full document, potentially leading to misinterpretations.

Recommendations

Users should be aware that this model is a screening tool, not an oracle. Any positive classifications should be reviewed by a qualified financial analyst to confirm their context and relevance. Do not rely on the model's output as the sole basis for financial or investment decisions.